Say it.
See it.
تكلّم… والباقي مُدوَّن
Modwen listens to your meetings in Arabic, English, or both in the same sentence, and turns them into notes, decisions and action items with real deadlines. No cloud. No account. Your voice never leaves your machine.
Decisions
Action items
Open questions
Everything happens inside your machine.
Recording, transcription, speaker detection and the AI that writes your notes all run on your own computer. Modwen refuses to talk to any server that isn't localhost. That rule is in the code, not in a policy.
Arabic, English, and everything in between.
Most meeting tools break when a sentence starts in Arabic and ends in English. Modwen was designed for exactly that.
Code-switching, natively
Whisper large-v3 with a per-sentence language rule keeps English terms in English and Arabic in Arabic: Gulf, Levantine, Egyptian and MSA.
Deadlines it really understands
"بكرة", "الخميس الجاي", "by end of day": resolved to real dates in your timezone. Invented deadlines are caught and dropped.
Notes that cite their source
Every decision and action item links to the exact moment it was said. Click it, and the transcript jumps there and plays the audio.
Your AI, your choice
Let Modwen download and run Qwen for you with one click, or point it at any local OpenAI-compatible server you already use.
Any meeting, one click
It captures your mic and the meeting audio on macOS and Windows, with live captions while you talk, and is ready for 2-hour meetings.
3 decisions · 5 action items
Stop. Get notified. Done.
Press Stop and keep working. A notification tells you when the summary, decisions, owners and open questions are ready.
Voice in. Anything out.
Meetings are just the start. Next, Modwen turns what you say into what you need, live: slides, charts, documents.
Up and running in three steps.
Download
Grab the latest release zip (about 0.5 MB) and unzip it anywhere.
Double-click
Modwen.command on macOS, Modwen.bat on Windows. The setup wizard picks models for your hardware and downloads them once.
Talk
Press Start in your next meeting. Press Stop. Read your notes.
# 1. download modwen-<version>.zip from Releases, unzip # 2. first time only: right-click → Open open Modwen.command # the setup wizard opens in your browser: choose "Download Qwen for me" # or connect your own local server, e.g. http://localhost:11434/v1 (Ollama)
:: 1. download modwen-<version>.zip from Releases, unzip :: 2. double-click (SmartScreen: More info → Run anyway) Modwen.bat :: NVIDIA GPU? CUDA acceleration is set up automatically.
git clone https://github.com/<you>/modwen && cd modwen uv sync --extra stt --extra desktop --extra mac # Windows: --extra windows (cd ui && npm install && npm run build) uv run modwen serve
Help Modwen understand every Arabic voice.
- Record a short dialect sample (with consent): Gulf, Levantine, Egyptian, Maghrebi.
- Test it on Windows and Linux, and tell us what breaks.
- Build the voice → slides layer with us.
Questions
Does anything really stay offline?
Yes. After the one-time model downloads, Modwen works with your internet turned off. AI endpoints must be localhost; anything else is refused in code.
What computer do I need?
16 GB of RAM minimum. Best: an Apple Silicon Mac with 32 GB, or a Windows PC with an NVIDIA GPU (12 GB+). Modwen detects your hardware and picks models that fit.
How good is the Arabic?
Transcription uses Whisper large-v3, which handles MSA well and Gulf, Levantine and Egyptian dialects increasingly well, including English words inside Arabic sentences. Summaries are best with Qwen3-8B or larger.
Can I use my own model?
Yes: any OpenAI-compatible server on your machine (llama.cpp, Ollama, LM Studio, vLLM). Switching is one setting.
Is it free?
Free and open source under Apache-2.0. No accounts, no subscriptions, no usage limits.